1. Field of the Invention
The present invention relates to an article take-out apparatus and article take-out method which recognize the positions and postures of articles stored in bulk in a 3D space and take out recognized articles using a robot.
2. Description of the Related Art
As this type of apparatus, in the past, there has been known an apparatus designed to recognize the positions of articles stored in bulk by carrying out pattern matching on a 2D image obtained by photographing the articles stored in bulk with a camera or on a 3D point set obtained by measurement with a 3D measuring device.
For example, Japanese Unexamined Patent Publication No. 2004-295223A (JP2004-295223A) describes an apparatus which prepares a 2D model pattern from a 2D image which is obtained by photographing an article in a reference 3D relative posture in advance, processes the 2D model pattern by 2D geometric conversion to prepare a plurality of converted 2D model patterns, and uses the plurality of converted 2D model patterns to carry out 2D pattern matching on a 2D image of an article.
Further, for example, the apparatus described in Japanese Unexamined Patent Publication No. 2011-179909A (JP2011-179909A) acquires a 3D model pattern of an article from a CAD model, etc. in advance while measuring the surface of the article in a 3D space by a 3D measuring device to acquire a 3D point set (distance image), and divides this 3D point set into sub-regions which are surrounded by edges extracted from the 3D point set. Further, first, one of the sub-regions is set as an article region. Processing for matching a 3D model pattern with this article region and processing for updating which adds another sub-region to the article region, that is, two processings, are repeated to measure the position and posture of an article.
However, the apparatuses which are described in JP2004-295223A and JP2011-179909A require the preparation of a 2D model pattern or 3D model pattern in advance for each type of article, so troublesome work is required. In particular, when there are a large number of types of articles, it is necessary to prepare that number's worth of model patterns, so tremendous work is required. Further, in the case of irregularly shaped articles, it is inherently not possible to create model patterns, so application is impossible.